If the sensor has been through span-calibration after production, a single-point calibration (20.9 %) at standard air should be enough. Source.
Can a 3-point calibration/cross-validation graph be used? Assuming medical O2 cylinders give purity ~ 97% (the 3rd point). Maybe these 3 points can test the oxygen sensor right off the box? Then, do a single-point test (20.9%) for most O2 analyzers. O2 Cylinder's availability should not be an issue?
Idea: